Image retrieval with relevance feedback

The proposed system for image retrieval using multidimensional features (IRMF) characterises and matches image content in a high dimensional feature space of colour, texture and shape dimensions. By including the entire pyramid of low-, medium-, and high-level primitives, the semantics of image content at different feature levels can be represented and extracted efficiently for image retrieval. This provides accurate query formulation and improves the accuracy in the search results. By co-jointly matching image features in a multidimensional space rather than in separate independent feature spaces, the precision in image retrieval is improved from more than 50% to up to 90% for the top 10 most similar images retrieved. The impact of the information of the image´s background has been mentioned in a very few published papers. Our experiments show that the efficient extraction of background information can improve the precision of image retrieval. To speed up the retrieval process, we also propose interactive relevance feedback to let the user participate in the process. The system is implemented for Internet Web access